System-based modeling is now widely used to quantify ecosystem and environmental elemental cycling (e.g., C and N), hydrological dynamics, and energy fluxes. In this context, deterministic differential equations link state variables and fluxes of ecosystems or environmental entities (e.g., lakes, forests, or areas of coastal ocean). Traditionally, these models are parameterized with limited observational data and then applied over extended temporal and spatial scales. These models are often structurally inadequate for representing the fundamental physical, chemical and biological processes, with parameter uncertainty, leading to divergence in system dynamics at coarse spatial and temporal scales.

In this collaborative project led by Prof. Zhuang of Purdue University and involving the MIT Joint Program and the Marine Biological Laboratory, we aim to transform the current system modeling approach by (1) Developing a stochastic version of the deterministic differential equation models of ecosystems and environmental systems; (2) Developing geospatial statistical techniques to fully exploit multifaceted observational data to improve model parameterization; (3) Developing advanced statistical and machine learning techniques to further utilize observational data to improve model structure; and (4) Applying the improved model to examine the economic implications compared with those obtained using the original modeling systems.

We will apply this novel approach to develop a cyber-enabled stochastic carbon-weather system to quantify net carbon exchanges between the terrestrial biosphere and atmosphere. Accurate quantification of regional carbon exchanges is critical to developing strategies for understanding carbon-climate-atmosphere feedbacks and the effects of pricing of greenhouse gas emissions and of land use change resulting from energy policy. Advantages of the proposed cyber-enabled terrestrial ecosystem model will include: (1) Quantification of regional simulations and associated uncertainty using new stochastic techniques and (2) Improved estimation of model parameters and structure using advanced statistical and machine learning techniques and spatiotemporal data acquired over ecosystems.

Project deliverables will include: (1) An innovative, cyber-enabled carbon-weather prediction system that can quantify net carbon exchanges between the biosphere and atmosphere and associated probabilistic information at high spatial and temporal resolution for the continental U.S., and (2) a suite of transformative advanced mathematical, statistical and system modeling techniques that could be applied to many other complex modeling problems (e.g. hydrological modeling). This project will significantly advance ecosystem sciences with computational thinking and will provide a unique opportunity to train a new generation of scientists in a highly interdisciplinary research environment.